Int. J. Electron. Commun.

(AE) 69 (2015) 699707

Contents lists available at ScienceDirect

International Journal of Electronics and

Communications (AE)
journal homepage: www.elsevier.com/locate/aeue

Transmission strategies and resource allocation for fading broadcast

relay channels
Arif Onder Isikman , Melda Yuksel
TOBB ETU, Turkey

a r t i c l e

i n f o

Article history:
Received 4 July 2014
Accepted 16 December 2014
Keywords:
Broadcast relay channel
Cooperation
Long-term power constraint
Outage probability
Power allocation

a b s t r a c t
In this paper the broadcast relay channel, where the source communicates with multiple destinations
with the help of a single relay is studied. Five different transmission protocols, direct transmission, multihop (MH), multihop with link combination (MHLC), path selection (PS) and path selection with link
combination (PSLC) are investigated. In MH and MHLC, the relay decodes the source message and assists
both destinations. In PS and PSLC, the relay can perform partial decoding and has the option to help only
one of the destinations. Under long-term power constraint, power allocation for delay-limited transmission is performed to minimize common outage probability and individual outage probability region. For
comparison, lower bounds on both common and individual outage probabilities are found. Numerical
results suggest that path selection signicantly lowers outage probabilities, while enforcing the relay to
help both destinations simultaneously is limiting the system performance.
2014 Elsevier GmbH. All rights reserved.

1. Introduction
The four-terminal broadcast relay channel (BRC), which is especially important for hierarchical downlink communications, is rst
studied in [1]. In the four-terminal BRC model, there are two destinations, which communicate with the source with the help of a
single dedicated relay. Various achievability schemes are proposed
for the BRC in [18]. The BRC can be especially important for nextgeneration wireless standards such as 3GPP Long Term Evolution
Advanced standard (LTE-A) [9,10], which proposes using relays for
enhanced data rates.
When channel state information is available at the transmitter,
substantial gains can be achieved if transmission rate and power
are adapted according to the channel conditions. Under long-term
power constraint over multiple fading blocks, there are several
adaptation techniques in the literature depending on the application:
If an application is delay tolerant, waterlling is the best strategy, which adjusts both power and rate over all fading blocks to
maximize the ergodic rate [11].
If an application is delay-limited, rate adaptation is not an
option. Instead there is a xed target transmission rate for each

Corresponding author. Tel.: +90 5389440004.

E-mail address: aisikman@etu.edu.tr (A.O. Isikman).
http://dx.doi.org/10.1016/j.aeue.2014.12.010
1434-8411/ 2014 Elsevier GmbH. All rights reserved.

transmission block, and only power is adjusted. If the target rate

is high, an outage is inevitable for some transmission blocks.
For such an application the optimal power allocation scheme
that minimizes outage probability is of threshold-type [12]. To
maintain a xed rate over time, channel inversion is performed.
However, it is best to cease transmission if too much power is
required to invert the channel.

In this paper, we are interested in delay-limited applications. Under long-term power constraint, optimal power allocation
protocols for minimum outage probability for fading broadcast
channels and relay channels are studied in [1316]. In [16], opportunistic protocols, in which the relay is not utilized if cooperation
consumes more power with respect to direct transmission are proposed and proved to perform close to the cut-set bound.
In this paper we investigate relaying strategies and related
power allocation methods that minimize outage probability for
the BRC with N destinations under long-term power constraint for
delay-limited applications. Note that, the four-terminal BRC is different from the two-receiver relay-broadcast channel [17,18]. In
the latter, the source communicates with two destinations one of
which acts as a relay for the other. The relaying receiver conveys
messages to a single node only and does not encounter the problem
of assisting two different receivers simultaneously.
In the single-antenna BRC under study, the source uses superposition coding at xed target rates to reach all destinations
reliably. We rst study the common outage probability for all of the

700

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

destinations. In a broadcast channel, there is common outage,

whenever any of the destinations is in outage [13]. Such an operation mode is necessary if coordination among receivers is to
be established and transmission to all of the destinations has to
take place at the same time. On the other hand, if coordination is
not required, destinations can declare outage independently. As a
result, for xed rates, each destination can have a different outage
probability, dening a region.
In the BRC setting, in addition to direct transmission (DT), we
investigate four different protocols: multihop (MH), multihop with
link combination (MHLC), path selection (PS) and path selection
with link combination (PSLC). We also upper bound the achievable
rates at the destinations and nd a lower bound on common and
individual outage probabilities and an upper bound on -outage
rate regions. In MH, MHLC, PS, and PSLC the source uses superposition coding for the N independent messages it has for each of
the N destinations and opportunistically resorts to DT, whenever
it consumes less power than using the relay. In MH, the relay has
to decode all N messages to help all destinations. We assume the
destinations only listen to the relay as this is a practical assumption that enables simple receivers. In PS; we exploit superposition
of messages, and allow the relay to assist a subset of destinations,
where the other destinations directly listen to the source. Similar to MH, in this case, the destinations that listen to the relay do
not listen to the source. Complementing MH and PS, in MHLC and
PSLC, we study the effect of link combination and explore the gains
introduced when the destinations can combine signals both from
the source and the relay. Although, MHLC and PSLC are sure to perform better than MH and PS, respectively, they require complex
receivers and their use is limited.
In our previous work [19], we only studied common outage
probability for the four-terminal BRC for DT, MH, PS and PSLC. In this
paper, we include MHLC and investigate individual outage probabilities. Moreover, in this work, the opportunistic behavior of the
source and the relay are determined optimally, whereas in [19], the
relay was utilized according to suboptimal rules. The contributions
of this work are listed as follows:
Under long-term power constraint and delay-limited transmission the minimum outage probability problem is posed for the
2-user BRC. Common and individual outage probability cases are
solved separately.
In addition to DT, four different opportunistic relaying strategies,
MH, MHLC, PS and PSLC are proposed and achievable rate regions
are given for the 2-user BRC. An upper bound on achievable rates
is also found. For these achievable schemes and the upper bound,
for xed target rates, optimal power allocation policies are determined that minimize the common outage probability and the
individual outage probability regions. The results are complemented with -outage rate regions, for a xed common outage
probability value and for a xed individual outage probability
vector.
The broadcast relay channel under study is not necessarily
degraded [2]. For some channel conditions it is better if both
destinations decode both messages, and for some other channel
conditions, less power is consumed if both destinations decode
their own messages only.
Computer simulations are performed for N = 2 and N = 3 to reveal
the effects of partial decoding at the relay for increasing number
of users. Our results indicate that PS is very close to optimal. We
also observe that partial decoding has signicant gains in the BRC
setting.
The organization of the rest of the paper is as follows. In
Section 2, system model and optimal power allocation principles
under common and individual outage constraints are introduced. In

Fig. 1. The broadcast relay channel (BRC), with one source (S), one relay (R), and
two destinations (D1 , D2 ). For the numerical simulations in Section 4, all four nodes
are located on a plane with SR, D1 -point P and D2 -point P distances are respectively
equal to d, d1 and d2 .

Section 3, transmission protocols are described in detail. In Section 4, numerical results are presented. Finally in Section 5, the
paper is concluded.
2. System model
The BRC consists of one source (S), one relay (R), and N destinations (D1 , D2 , . . ., DN ). The model is illustrated for N = 2 in Fig. 1. The
instantaneous amplitude squares of complex channel gains among
SDj , SR, and RDj are respectively denoted by aj , b, and cj , j = 1,
2, . . ., N. It is assumed that the channel gain amplitudes are known
globally at all nodes, whereas channel gain phases are known only
at corresponding receivers. The channel coefcients have quasistatic fading [12] and are independent from one block to the other.
Complex Gaussian noise at the receivers are independent, and have
zero mean and unit variance.
We assume that the relay is half-duplex and there is time division among the source and the relay. The source transmits for t
fraction of the block, 0 < t 1, and the relay transmits in the rest
1 t.
The system is delay-limited. Communication lasts for B blocks,
where B approaches innity. Over each communication block, the
source transmits N independent messages W1 , . . . WN at xed target
rates R1 , . . ., RN respectively to each destination. The source node
encodes W1 , . . ., WN into X1 , . . ., XN using superposition coding [20].
(i)
It allocates power PSj (s, t) to send Xj , j = 1, . . ., N, and transmits
X = X1 + X2 + + XN to reach all destinations simultaneously. Here
s = (a1 , . . ., aN , b, c1 , . . ., cN ) is the channel state vector and i, i = DT,
MH, MHLC, PS, PSLC, denotes the transmission protocol that will
be dened in the next subsection. Upon receiving the source signal
X, the relay decodes Wj for all j E(s), where E(s) {1, 2, . . ., N}.
Note that the subset E(s) depends on the channel state s, and on the
transmission protocol. The relay then reencodes Wj , j E(s) into
(i)
X j using an independent codebook at power P (s, t) and transmits

Rj

X .
the sum
jE(s) j
Before explaining power and outage constraints, we next outline
the transmission protocols, DT, MH, MHLC, PS and PSLC. Calculation
of the instantaneous achievable rates at the destinations and the
required power levels will be presented in Section 3.
2.1. Overview of transmission protocols
In DT, the relay is not utilized and the system is equivalent to a
broadcast channel. The source node transmits all the time, t = 1.
Note that, DT will be a part of all other opportunistic protocols

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

701

any of these two constraints is relaxed, MH and MHLC respectively

become equivalent to PS and PSLC. Therefore, for MH and MHLC,
we do not study the individual outage probability region.
(i)
In Section 3, we write achievable rates at each destination Cj for
N = 2 case only for enhanced readability. We nd the instantaneous
(i)
achievable rates at each destination Cj , j = 1, 2, i =DT, MH, MHLC,
PS, PSLC and an upper bound on achievable rates for each channel
state s and for a xed t. We will present results for N = 3 in Section 4.
In the following, we explain power and outage constraints.

2.2. Long-term average total power constraint

Fig. 2. (a) Direct transmission operation modes: for common outage, only DT is possible, whereas for individual outage the system can be in DT, DT1 or DT2 modes.
(b) Operation modes for MH and MHLC for common outage probability: direct
transmission (DT), decode-and-forward (DF) and decode-and-forward with link
combination (DFLC). In MH, DT and DF modes are available, while in MHLC DT and
DFLC modes are in use.

under study, in which the relay is not employed, whenever it results

in less consumption. Fig. 2a represents the operation modes for
DT for N = 2. Under common outage constraint, there is only one
operation mode: The source node transmits to both destinations
simultaneously. With a slight abuse of notation, this mode is also
denoted as DT. Under individual outage constraint, there are three
operation modes: The source can serve D1 only (DT1), D2 only (DT2)
or D1 and D2 together (DT). When
  generalized to N destinations,
N
N
there will be a total of
= 2N 1 operation modes for
n=1
n
individual outage.
The two modes of operation for MH, direct transmission (DT)
and decode-and-forward (DF), for the four-terminal BRC for common outage probability are shown in Fig. 2b. For N destinations
under common outage DT and DF are still the only mode options.
The relay behaves opportunistically and chooses the less power
consuming mode over the other. In MH the destinations only listen to the relay in DF mode. In MHLC, the two modes of operation
are DT and decode-and-forward with link combination (DFLC). In
MHLC protocol in DFLC mode, the destinations listen to the signals
from both the source and the relay combining the incoming single
and double links in Fig. 2b.
In PS, the source transmits to all destinations. However, the relay
is not required to decode all messages, but can decode any subset
of them. This decreases the decoding constraint at the relay, and
increases its chance to be more useful for the destinations. Specically, for the four-terminal BRC for common outage the four modes
of operation are DT, relay helps only D1 (RH1), relay helps only D2
(RH2) and DF. These additional operation modes, RH1 and RH2 are
shown in Fig. 3a. In addition to these operation modes, 4 additional
cases appear when individual outage declaration is allowed. The
source can transmit to D1 individually either directly (DT1) or via
the relay (DF1). Similarly, the source can transmit to D2 in DT2 or
DF2 modes. DF1 and DF2 are shown in Fig. 3a. Then, for general N
N
there are
 2 operation modes for common outage, and an additional
N1 N
2n number of modes for individual outage.
n=1
n
In PSLC, for the four-terminal BRC under common outage the
four modes of operation are DT, relay helps only D1 with link combination (RH1LC), relay helps only D2 with link combination (RH2LC)
and DFLC. Under individual outage, in addition to the operation
modes DT DFLC, RH1LC and RH2LC, there are DF1LC and DF2LC
modes shown in Fig. 3b.
In MH and MHLC protocols, the source sends both messages W1
and W2 . Then the relay decodes and forwards both messages. When

We assume the source and the relay have a long-term average

total power constraint Pavg over all communication blocks:


(i)

(i)

[t PS (s, t) + (1 t) PR (s, t)] f (s) ds Pavg

(1)

where
(i)

(i)

(i)

(i)

(2)

(i)

(i)

(i)

(i)

(3)

PS (s, t) = PS1 (s, t) + PS2 (s, t) + + PSN (s, t)

PR (s, t) = PR1 (s, t) + PR2 (s, t) + + PRN (s, t)

and f(s) denotes the probability density function of the channel

state s.
(i)
Let Cj (s, t) denote the instantaneous achievable rate at the
jth destination j = 1, 2, . . ., N with protocol i, i = DT, MH, MHLC,
PS, PSLC, for a xed t and channel state vector s. For reliable communication, we require the instantaneous achievable
(i)
rates Cj (s, t), which will be calculated for all destinations and
for all protocols in Section 3, to be larger than the xed target
rates Rj as the system is assumed to be delay-limited. Den(i)
(i)
(i)
(i)
ing (P (s, t), . . ., P (s, t), P (s, t), . . ., P (s, t)) as the minimum
S1

SN

R1

RN

(i)

amount of power levels that satisfy the condition Cj (s, t) Rj , the

minimum total amount of power required for protocol i,1 at state s
is given as
(i)
(i)
(i)
(i)
Preq (s) = mint(P S1 (s, t) + + P SN (s, t)) + t (P R1 (s, t) +
t

(i)

+ P RN (s, t)),

(4)

(i)
(i)
(i)
(i)
= t (P S1 (s, t ) + + P SN (s, t )) + t (P R1 (s, t ) + + P RN (s, t )),

(5)
where t* is the best t, 0 < t 1 that minimizes (4), t = 1 t, and t =
1 t . Note that the fraction of the time the relay listens, t* , is a
function of the channel state vector s.

2.3. Common outage probability

The common outage probability is introduced in [13]. In a broadcast channel, depending on the channel state, the broadcast channel
can either not be used at all, or transmission to all the receivers
take place at the same time. In other words, whenever one of the
destinations is in outage, there is system outage.
(i)
Given Preq (s) for the ith protocol at state s, the optimal resource
allocation function that attains the minimum common outage

702

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

probability problem of (7) and the -outage rate region problem

are inherently the same [13].
2.4. Individual outage probability region
In a broadcast channel, instead of declaring system outage, every
destination can go into outage independently from each other.
Then, there are N different outage probabilities for each user. These
(i)
(i)
probabilities will be denoted as Pout,1 , . . ., Pout,N where i = DT, MH,
(i)

Fig. 3. (a) Additional operation modes for PS and PSLC for common outage: relay
helps user 1 (RH1), relay helps user 1 with link combination (RH1LC), relay helps
user 2 (RH2) and relay helps user 2 with link combination (RH2LC). Then, in PS, for
common outage the operation modes are DT, RH1, RH2 and DF. Similarly, in PSLC,
DT, RH1LC, RH2LC and DFLC modes are possible for common outage. (b) Additional
operation modes for individual outage probability: decode-and-forward 1 (DF1),
decode-and-forward 1 with link combination (DF1LC), decode-and-forward 2 (DF2),
and decode-and-forward 2 with link combination (DF2LC). Then, in PS, for individual
outage the operation modes are DT, RH1, RH2, DF, DT1, DF1, DT2 and DF2. Similarly,
in PSLC, DT, RH1LC, RH2LC, DFLC, DT1, DF1LC, DT2 and DF2LC modes are possible
for individual outage.

probability is of threshold type [12,13,16]. As a result, the required

power levels are determined according to

(i)
Pkj (s, t )

(i)

Preq (s) Pth

if

(i)
P kj (s, t )

if Preq (s) < Pth ,

(i)

(6)

for k = S, R, j = 1, 2, . . ., N where Pth is determined such that (1) is

satised.2
This threshold type behavior can be explained as follows: As
the application is delay-limited, reliable communication is needed
for each channel state s. Then, one would choose the transmis(i)
sion power levels such that the instantaneous achievable rate Cj
is equal to the target rate Rj : There is no point in achieving more
than required. However, for very poor channel conditions, the total
power level required to send the target rates R1 , . . ., RN respectively
to D1 , . . ., DN is very high. If transmission is sustained during such
poor channel conditions, power is wasted to invert the channel
and the total average power constraint in (1) is violated. However, if transmission is discontinued until the channel conditions
become favorable, then power can be saved and (1) is satised.
For those channel states during which transmission is off, or when
(i)
Preq (s) Pth , the system is in outage.
Subject to the total average power constraint Pavg , one can determine Pth such that the common outage probability for protocol i,
(i)
(i)
Pout = P(Preq (s) Pth ), is minimized. Then for a given target rate
vector (R1 , . . ., RN ), the minimum common outage probability is
given as
(i)

(i)

Pout-min = minPout
Pth

subject to(1).

(7)

In this paper our rst objective is to nd the minimum common

outage probability dened in (7). We investigate ve protocols DT,
MH, MHLC, PS, and PSLC. In each of these protocols, all destinations
are turned on or off simultaneously. In addition to minimum common outage probability for a given rate vector (R1 , . . ., RN ), we also
analyze the -outage rate region. The -outage rate region is the
collection of all achievable rate vectors (R1 , . . ., RN ), for which the
common outage probability is at most  and the total average power
constraint (1) is satised. Note that the minimum common outage

(i)

MHLC, PS, PSLC. The collection of all achievable (Pout,1 , . . ., Pout,N )

for a given power allocation and for a given target rate vector (R1 ,
. . ., RN ) denes the individual outage probability region. In this
paper our second objective is to nd the individual outage probability region and the corresponding -outage rate region for the
transmission strategies DT, MH, MHLC, PS and PSLC. Note that the
individual outage measure does not reduce the system to the classical relay channel with a single destination, as in the BRC there
is still the option of broadcasting to all or a subset of destinations
simultaneously.
Under the individual outage probability measure, for each channel state s, the source can choose to transmit to a subset of the
destinations resulting in 2N 1 possible sets of active users. Note
that this is not an option under common outage, in which the system automatically goes into outage whenever the source does not
transmit to all destinations simultaneously. We will denote the
active user set with Sl , l = 1, 2, . . ., 2N 1 and the required power
(i)
level for the active user set Sl with Preq,l (s) for each protocol i for
state s. To nd the best active user set for each channel state s, we
follow the algorithm described in [13]. Due to lack of space, we
leave the details to the reader.
3. Transmission protocols
In this section we describe DT and the opportunistic protocols MH, MHLC, PS and PSLC in detail and also derive performance
bounds on outage probability and -outage rate regions.
3.1. Direct transmission
The minimum common outage probability and the individual
outage probability region for the broadcast channel for DT is solved
in [13]. Here, we restate the results, as DT is a part of all other
opportunistic protocols under study.
3.1.1. Common outage
When the performance metric is system outage, the best
strategy for the source is to communicate with all destinations
simultaneously. The source can only consume more power for the
same outage probability if it chooses to transmit only a subset of
the messages.
Let () be a permutation on {1, 2} such that a(1) > a(2) . Then
(DT)

(DT)

the point (C1 , C2 ) is on the capacity region of the broadcast

channel for state s, where

(DT)
C(1)

= log

(DT)
1 + a(1) PS(1)


,

(DT)
C(2)

= log

1+

(DT)

a(2) PS(2)
(DT)

1 + a(2) PS(1)

.
(8)

(DT)

(i)

Preq (s) notation will also be used for operation modes DT1, DT2, DF, DF1, DF2,
RH1, RH2, RH1LC, RH2LC, DFLC, DF1LC and DF2LC.
(i)
(i)
(i)
(i)
2
From this point on, we will denote P kj (s, t), Pkj (s, t), Pk (s, t) and Cj (s, t) respec(i)

(i)

(i)

(i)

tively with P kj , Pkj , Pk and Cj for a simpler notation.

Solving C(l) = R(l) for PS(l) , l = 1, 2, one can calculate the minimum
(DT)
amount of power required, P S(l) , to attain the xed target rates, R1

and R2 . In the following we nd the required source powers for DT

for the four-terminal BRC.

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

Dene g as a function with three inputs and two outputs as

(y1 , y2 )  g(x1 , x2 , x3 )

R1

2 x3

y1 =
R1

2 x3

R2

2 x3

y2 =
R2

2 x3

(9)

1
x1

1
x1

1
x2

if x1 x2

+ y2

+ y1

1
x2

S1

(DF)

(DF)

C(1) = t log 1 + c(1) PR(1) ,

(DF)
C(2)

if x1 x2

= t log

S2

1+

(DF)

1 + c(2) PR(1)

(10)

Using (10),(6) and (7) one can nd the minimum common outage
probability.
3.1.2. Individual outage
Unlike the common outage case, when there is individual outage, the source no longer has to serve both destinations at the same
time. Depending on the channel state, communicating with only
one of the destinations can be more power efcient in the longterm. When there is an individual outage constraint, there are 3
possible sets of active users, and thus 3 modes of operation: DT1,
DT2, and DT, shown in Fig. 2a. In DT1 and DT2 modes, the system
turns into a point-to-point channel as the source communicates
with only one of the destinations. The achievable rate for DT1 mode
(DT 1)
(DT1)
(DT1)
is given by C1
= log(1 + a1 PS1 ). Then, solving C1
= R1 , we
(DT1)
nd P
as
S1

2R1 1
.
a1

(11)

(DT1)
(DT1)
(DT1)
= 0. As the relay is not utilized, P R1
= P R2
= 0 and
while P S2
(DT)
(DT1)
(DT1)
(DT)

thus P
(s) = Preq (s) = P
is obtained. Similarly, P
(s) can
req,1

req,2

S1

be found for DT2 mode. Finally, when both users are active, the system is equivalent to the broadcast channel and the required power
(DT)
(DT)
(DT)
(DT)
levels are the same as P S1 and P S2 resulting in Preq,3 (s) = Preq (s)
(DT)

where Preq (s) computed for the common outage case in (10). Given
(DT)
Preq,1 (s),

(DT)

(DT)

Preq,2 (s) and Preq,3 (s), one can determine the best active
user set and nd the individual outage probabilities as in [13], as
described in Section 2.4.

3.2. Multihop
For MH, we only consider common outage. As mentioned before,
there are 2 modes of operation in MH: DT and DF. In DF mode, the
relay has to decode both messages, W1 and W2 .
(DF)
The achievable sum rate for W1 and W2 at the relay, CR (s),
is found as

(DF)
CR

= t log

(DF)
1 + bPS

(DF)

(DF)

(DF)
= (2
P S

(14)
(DF)

Then substituting (12) and (14) into (4), we obtain Preq (s). As MH

(MH)

Preq (s) = min

(DF)

(MH)
Preq (s)

(DT)

Preq (s), Preq (s)

is calculated as

(15)

(DF)

where Preq (s) is given in (10). Using (15) and (6), the minimum
common outage probability of (7) is calculated.
3.3. Multihop with link combination
In MHLC, under common outage the system is either in DT or
DFLC modes. In DFLC mode, the destinations listen to both the
source and the relay. The source uses superposition coding and
sends W1 and W2 while the relay has to decode and forward
(DFLC)
both W1 and W2 . Thus P S
(s) is given by (12). Upon successful
decoding, the relay forwards both messages to both destinations.
The destinations have two distinct observations about (W1 , W2 ).
However, the BRC under study is not necessarily degraded as
dened in [2]. Note that in a degraded Gaussian broadcast channel
with two receivers, superposition coding with successive cancellation decoding is optimal [2]. While the user with worse channel
conditions decodes its own message only, the user with better
channel conditions decodes its own message after decoding the
other users message. However in a BRC we cannot conclude that
one of the destinations can always decode the other users message without incurring an extra cost. Therefore, we study both
cases: a destination i can decode its message Wi either directly
(treating Xj , i =
/ j as noise) or after decoding the other users
message.
3.3.1. Case 1
Both destinations decode their messages directly, i.e., D1 directly
decodes W1 and D2 directly decodes W2 . Then, the achievable rates
(DFLC-1)
(DFLC-1)
at the destinations are C1
= K1 and C2
= K2 , where

K1 = t log

1+

(DFLC)

a1 PS1

+ (1 t) log

(DFLC)

1 + a1 PS2

1+

(DFLC)

c1 PR1

(DFLC)

1 + c1 PR2

(16)

. In DF mode, in order for the

(DF)

1
1) .
b

(13)

relay to decode both messages reliably, CR

the target sum rate R1 + R2 . Thus,
R1 +R2
t

(P R1 , P R2 ) = g(c1 , c2 , 1 t).

is an opportunistic protocol,

(DT)
(DT)
(DT)
Preq (s) = P S1 + P S2 .

(DT 1)
P S1
=

(DF)

c(2) PR(2)

where t = 1 t and the relay power needed for reliable reception at

(DF)
(DF)
the destinations can be found solving C(1) = R(1) and C(2) = R(2)
as

if x1 < x2

(DT)
(DT)
Then, P S1 and P S2 can be expressed in terms of the function g
(DT) (DT)
(DT)

as (P
,P
) = g(a1 , a2 , 1). Then, Preq (s) can be calculated from

(5) as

Using independent codebooks, the relay then reencodes W1 and

W2 , and forwards them to the destinations using superposition
coding. As the destinations only listen to relay and the source and
the relay transmissions are orthogonal, the transmission from the
relay to the destinations constitute a broadcast channel. Let () be
a permutation on {1, 2} such that c(1) > c(2) . Then the achievable
rates at the destinations are


if x1 < x2

703

must be as large as
K2 = t log
(12)

1+

(DFLC)

a2 PS2

(DFLC)

1 + a2 PS1

+ (1 t) log

1+

(DFLC)

c2 PR2

(DFLC)

1 + c2 PR1

(17)

704

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

3.3.2. Case 2
D1 rst decodes W2 . Subtracting X2 and X 2 from its received signals respectively in the rst and the second slots, D1 then decodes
W1 . D2 , on the other hand, directly decodes W2 . As W2 has to be
decoded reliably at both destinations, the achievable rate for W2 is
(DFLC-2)
= min{K2 , L2 } where K2 is dened in (17) and L2 is given
C2
by

L2 = t log

(DFLC)

1+

a1 PS2

+ (1 t) log

(DFLC)

1 + a1 PS1

(DFLC)

1+

c1 PR2

(DFLC)

1 + c1 PR1

(18)
Once the effect of W2 is removed from D1 s observations, D1 can
(DFLC-2)
attain the rate C1
= M1 where M1 is given by
M1 = t log

(DFLC)
1 + a1 PS1

+ (1 t) log

(DFLC)
1 + c1 PR1

(19)

3.3.3. Case 3
Case 3 complements Case 2. D1 directly decodes W1 while D2
rst decodes W1 and then decodes W2 . In this case, W1 has to
be decoded reliably at both destinations. The constraints on the
(DFLC-3)
achievable rate at D1 are given by C1
= min{K1 , L1 } where K1
is dened in (16) and L1 is given by

L1 = t log

1+

(DFLC)

a2 PS1

(DFLC)
1 + a2 PS2

+ (1 t) log

(DFLC)

1+

c2 PR1

(DFLC)
1 + c2 PR2

3.4. Path selection

In the PS protocol, the four modes of operation for the fourterminal BRC are DT, RH1, RH2 and DF.
(RH1)
In RH1, the source transmits W1 to the relay with power P S1
(RH1)
and W2 to D2 with power P
using superposition coding. In other
S2

words, the transmission from S to R and D2 forms a broadcast channel. In RH1, D1 does not listen to the source, but only to the relay
and the relay is only required to decode W1 . In RH1 mode, if b > a2 ,
the source superimposes X1 on X2 , and the opposite is true if b < a2 .
Then the achievable rates at the relay is




(RH1)

t log 1 + bPS1
ifb > a2

(RH1)

CR
=
(RH1)
bPS1

otherwise
t log 1 +
(RH1)
1 + bPS2

Note that in RH1, D2 does not need to listen to the relay, as the
relays transmission carries information only about W1 and is of
no use at D2 . Upon successfully decoding W1 , the relay independently reencodes and forwards W1 to D1 . The achievable rates at
the destinations are
(RH1)

C1

(RH1)

C2

(DFLC)

(DFLC)

+ (1 t) log 1 + c2 PR2

(21)

3.3.4. Case 4
In Case 4, both receivers decode both messages. Then, the
(DFLC-4)
= min{L1 , M1 }
achievable rates at D1 and D2 are given by C1
(DFLC-4)
C2

and
= min{L2 , M2 }, respectively.
There is no guarantee that Case 2 or Case 3 achieves the minimum total required power in a communication block. Therefore,
all 4 cases need to be considered.
(DFLC)
As in MH, for reliable decoding at the relay, CR
must be
(DFLC)
is the
as large as R1 + R2 . Thus, the required source power P
S

same as (12). Note that the superposition order at the source does
(DFLC)
not affect the total source power P S
, but the individual power
(DFLC)
(DFLC)

levels P
and P
have an effect on the relay power levels
S1

S2

that will be calculated shortly. Let [0, 1] denote the fraction of

(DFLC)
source power which is allocated to transmit W1 . Then, P S1
=
(DFLC)
(DFLC)
(DFLC)

P
,P
= (1 )P
.
S

S2

(DFLC)
(DFLC)
Substituting the source power levels, P S1
, and P S2
, into

(RH1)

t log 1 +

(24)

(RH1)

a2 PS2

ifb > a2

(RH1)

1 + a2 PS1

(25)




t log 1 + a2 P (RH1)

otherwise

S2

Then, removing the effect of W1 from its observations, D2 can

(DFLC-3)
achieve the rate C2
= M2 where M2 is given by
M2 = t log 1 + a2 PS2

= (1 t) log 1 + c1 PR1

.
(20)

(23)

3.4.1. Common outage

(RH1)
The required power levels for the source in RH1, P S1
and
(RH1)
(RH1)
(RH1)
P
are found by solving C
= R1 and C
= R2 and given
R

S2

(RH1)
(RH1)
as (P SR , P S2 ) = g(b, a2 , t). The required power level for the
(RH1)
(RH1)
(RH1)
relay P
is found by solving C
= R1 and given as P
=
R

R1

R1

1
(RH1)
(RH1) (RH1) (RH1)
= 0. Substituting P S1 , P S2 , P R1 and
(2 1t 1)/c1 while P R2
(RH1)
(RH1)
P R2
into (5), we obtain Preq (s).
The RH2 mode is similar to RH1 and the relay only assists D2 .
In RH2, the required source and relay power levels are given as
R2
(RH2)
(RH2)
(RH2)
(RH2)
(P S1 , P S2 ) = g(a1 , b, t) and (P R1 , P R2 ) = [0, (2 1t 1)/c2 ].

(DF)

The required power for DF is the same as in MH, Preq (s) is calculated using (12) and (14). Then the required power level for PS is
given as
(PS)

Preq (s) = min

(DF)

(RH1)

Preq (s), Preq

(RH2)

(s), Preq

(DT)

(s), Preq (s)

(26)

which can then be used to calculate the minimum common outage

probability of (7).

(DFLC-n)

(16)(21) and satisfying Cm

= Rm , m = 1, 2, n = 1, 2, 3, 4, the
destinations nd the optimal case and calculate the required relay
(DFLC)
(DFLC)
power levels P R1
and P R2
and optimal
that minimize
the total required power for a communication block. Substituting
(DFLC) (DFLC) (DFLC)
(DFLC)
P S1
, PS2
, PR1
and P R2
into (5), we obtain the required
(DFLC)

power level for DFLC denoted as Preq

(MHLC)

Preq

(s) = min

(DFLC)

Preq

(DT)

(s). Finally,

(s), Preq (s)

(22)

and the common outage probability of (7) can be computed using

(22) and (6).

3.4.2. Individual outage

In PS, under individual outage constraint, in addition to the operation modes DT, RH1, RH2, and DF, there are four operation modes
DT1, DT2, DF1 and DF2, as shown in Fig. 3b.
If only D1 is active, then the possible operation modes are DT1
and DF1. In DT1, the source directly sends W1 to D1 without the help
(DT1)
of the relay. The required power for reliable transmission Preq is
equal to (11).
In DF1 mode, the source sends W1 to D1 via the relay. The achievable rates at the relay and the destination are respectively given
by

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

(DF1)

CR

(DF1)

= t log 1 + bP S1

(DF1)

C1

(DF1)

= (1 t) log 1 + c1 PR1

3.6. Performance bounds


.

(27)
(DF1)
CR
= R1 and
(DF1)
thus Preq (s) are obtained.
(DT1)
(DF1)
min{Preq (s), Preq (s)}.

(DF1)

Solving

C1
As

= R1 ,

(DF1)
P S2

(DF1)
P S1 ,
(DF1)
P R2 =

(DF1)
P R1

705

and

(PS)

0, Preq,1 (s) =

Replacing the subscripts 1 with 2 and following the same

(PS)
steps as above Preq,2 (s) can be found for DF2.
Finally, when both users are active, the required power
levels are the same under individual and common outage con(PS)
(PS)
(PS)
(PS)
straints. Then, Preq,3 (s) = Preq (s) in (26). Given Preq,1 , Preq,2 and

In the following, we write upper bounds on the achievable rates

in the four-terminal BRC. Using these, we deduce a lower bound on
outage probability and an upper bound on -outage rate region.
We derive two different upper bounds on the achievable rates
at the destinations. In the rst bound, we assume the relay is
given the source messages W1 and W2 for free, or equivalently
the relay can always decode W1 and W2 reliably. Then the system becomes equivalent to parallel broadcast channels, for which
the capacity region is given in [21]. For a xed t, achievable rates
at the destinations are respectively upper bounded by Cj , where
(DFLC)

(DFLC)

Preq,3 we nd the individual outage probabilities as described

in Section 2.4.

Cj = Cj
, j = 1, 2 and Cj
are calculated as in Section 3.3.
The multiple antenna broadcast channel with two antennas at
each destination and a single antenna at the source constitutes the
second upper bound on the BRC under study [7]. Then, the achievable rates at the destinations for a xed t are upper bounded with

3.5. Path selection with link combination

(bound)
C (1) = t log 1 + (a(1) + b)PS(1)

(PS)

In PSLC, the operation modes are DT, RH1LC, RH2LC and DFLC. In
PSLC, the destinations always listen to the source. Then, in RH1LC
(RH1LC)
mode the achievable rate at Dj , Cj
is calculated similar to
MHLC as described in Section 3.3. Setting the channel gain magni(RH1LC)
tude square c2 = 0, the power level PR2
= 0 and replacing DFLC
(RH1LC-n)

with RH1LC in (16)(21), Cj

for all cases.

, j = 1, 2, n = 1, 2, 3, 4 is found

3.5.1. Common outage

(RH1LC) (RH1LC)
In PSLC, in RH1LC, the source sets (P S1
, PS2
) = g(b, a2 , t)
(RH1LC)
(RH1LC)
as in the RH1 mode of PS. Substituting P
and P
in
S1

(RH1LC-n)

S2

= R1 , n = 1, 2, 3, 4, D1 nds the
(RH1LC)
for
optimal case and calculates the required relay power P R1
reliable reception when there is link combination. As the relay
(RH1LC)
(RH1LC) (RH1LC) (RH1LC)
does not help D2 , P R2
= 0. Using P S1
, PS2
, PR1
and
(RH1LC)
(RH1LC)
P
, we obtain Preq
(s). The required power levels for RH2LC

(16)(21) and satisfying C1

R2

(RH2LC)

mode, Preq

(s), are similarly obtained.

In DFLC, the required power level,

(PSLC)

DFLC in MHLC. Therefore, Preq

(PSLC)

Preq

(DT)

(RH1LC)

(s) = min{Preq (s), Preq

(DFLC)
Preq (s),

is the same with

(s) is given as
(RH2LC)

(s), Preq

(DFLC)

(s), Preq

(s)}.

(28)

We then use (28) and (6) to compute the minimum common outage
probability of (7).

C (2) = t log

(31)

(bound)

1+

(a(2) + b)PS(2)

(bound)

1 + (a(2) + b)PS(1)

where the permutation (l) is a permutation on {1, 2} such that

a(1) > a(2) . Combining the two bounds, achievable rates at D1 and
D2 are respectively upper bounded by min(C1 , C1 ) and min(C2 , C2 ).
3.6.1. Common outage
Solving min(C1 , C1 ) = R1 and min(C2 , C2 ) = R2 , we calculate
(bound)
(bound)
(bound)
(bound)
(bound)
P
(s), P
(s), P
(s) and P
(s) to obtain Preq
(s).
S1

S2

(bound)

R1

R2

Then we use Preq

(s) to obtain the minimum common outage
probability dened in (7). This minimum common outage prob(bound)
ability Pout-min is a lower bound on all other protocols DT, MH,
MHLC, PS and PSLC. Based on the upper bounds on achievable rates,
min(C1 , C1 ) and min(C2 , C2 ), we also compute an upper bound on
the -outage rate region in the next section.
3.6.2. Individual outage
The bounds calculated in Section 3.6.1 are useful when both D1
and D2 are active. However, when there is individual outage, it is
possible that only one of the destinations is active. In that case, BRC
reduces to the three terminal relay channel [22]. Using the cut-set
bounds for the relay channel [16,20] we can write two additional
bounds on the achievable rate at Dj , j = 1, 2, as

(bound)
(bound)
) + (1 t) log(1 + cj PRj
C j = t log 1 + aj PSj

3.5.2. Individual outage

In PSLC, if only D1 is active, the system can either be in DT1 or
DF1LC modes. When compared with PS, the only change in required
power levels occur in DF1LC mode, for which the achievable rate at
D1 now becomes
(DF1LC)

C1

(DF1LC)

= t log 1 + a1 PS1

(DF1LC)

+ (1 t) log 1 + c1 PR1


.
(29)

(DF1)
CR

(DF1LC)
C1

Solving
= R1 and
(DF1LC)
(DF1LC)
P
(s), and Preq
(s). Then

= R1 , we obtain

(DF1LC)
P S1
(s),

R1

(PSLC)

(DT1)

(DF1LC)

Preq,1 (s) = min{Preq (s), Preq

(PSLC)

(s)}.

(30)
(PSLC)

(PSLC)

Similarly, Preq,2 (s) is obtained. Finally, Preq,3 (s) = Preq (s) in (28).
Then, the individual outage probabilities are found as explained in
Section 2.4.

(32)

(33)

(bound)
C j = t log 1 + (aj + b)PSj
.
(bound)

Solving min(C j , C j ) = Rj , Preq,j

(bound)
Preq,3 (s)

(34)
(s) can be obtained. Moreover,

(bound)
Preq
(s),

=
where the latter is found in Section 3.6.1.
As a result, the individual outage probabilities are computed.

4. Numerical results
For the numerical results, we will investigate N = 2 and N = 3
cases. We assume a1 , a2 , a3 , b, c1 , c2 , c3 are independent exponential
random variables. For N = 2 we assume all terminals are located on
a plane. The relay is located on the line joining the source and point
P, where SR distance is d and the distance between the source
and point P is normalized to 1. In Fig. 1 Dj P distance is denoted
as dj , j = 1, 2. Similarly, for general BRC with N destinations, each
destination is located on the plane which is orthogonal to the SP

706

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

Fig. 4. (a) The minimum common outage probability vs. average total power, N = 2, R1 = 1, R2 = 1, d = 0.2, d1 = 0.25, d2 = 0.25, = 4. (b) The minimum common outage probability
vs. average total power, N = 3, R1 = 1, R2 = 1, R3 = 1, d = 0.2, d1 = 0.25, d2 = 0.25, d3 = 0.25 and = 4.

Fig. 5. (a) The -outage rate region for xed minimum common outage probability of 0.01, N = 2, Pavg = 1 dB, d = 0.3, d1 = 0.2, d2 = 0.4 and = 4. (b) The -outage region for
xed minimum common outage probability of 0.01 and for xed individual outage probability Pout,1 = Pout,2 = 0.01, N = 2, Pavg = 1 dB, d = 0.3, d1 = 0.2, d2 = 0.4 and = 4.

Fig. 6. (a) E{} and E{} (dened in (35)) vs. d2 for N = 2, Pavg = 0.5 dB, R1 = 1, R2 = 1, d1 = 0.5 and = 4 for PSLC. (b) Individual outage probability region for N = 2, Pavg = 4 dB,
d = 0.2, d1 = 0.25, d2 = 0.25, R1 = 1, R2 = 1, and = 4.

line in three dimensional space with dj , j = 1, 2, . . ., N representing Dj P distance. As a result, the random variables aj , b, and cj ,
j = 1, 2, . . ., N respectively have the mean values (1 + dj2 )
2

, d

and [(1 d) + dj2 ] , where is the path loss exponent.

Fig. 4a illustrates the minimum common outage probability vs.
average total power for all protocols for N = 2 with R1 = 1, R2 = 1,
d = 0.2, d1 = 0.25, d2 = 0.25 and = 4. We observe that for the delaylimited system under study, MH requires approximately 4 dB less
total average power with respect to DT at Pout = 101 . However, it
is far from optimal. Link combination, MHLC, adds 0.3 dB gain upon
MH. On the other hand, PS signicantly improves upon MH and is
only 0.2 dB away from the lower bound. This shows that enforcing

the relay to help both destinations simultaneously is quite limiting,

and path selection is necessary. As PS performs very close to PSLC,
we can say that the gains obtained are mainly due to path selection
rather than link combination at the destinations when d = 0.2. This
implies that simple receivers are sufcient when the relay is close
to the source. In the simulations it is observed that in MHLC and
PSLC in DFLC mode, in approximately 1% of all channel states the
system is either in Case 1 or Case 4 as described in Section 3.3. We
conclude that in the resultant BRC implementing the given transmission schemes, one of the receivers does not always have a better
observation than the other. Although rarely, it can be benecial if
both receivers decode both messages, or both receivers decode only
their own.

A.O. Isikman, M. Yuksel / Int. J. Electron. Commun. (AE) 69 (2015) 699707

Fig. 4b illustrates the minimum common outage probability vs.

average total power for three users for all protocols for R1 = 1, R2 = 1,
R3 = 1, d = 0.2, d1 = 0.25, d2 = 0.25, d3 = 0.25 and = 4. When compared with Fig. 4a, we observe that all protocols and the lower
bound require more power to attain the same common outage
probability for an additional user. This loss is approximately 3 dB
for the lower bound, PS and PSLC, and 4 dB for MH and MHLC. Moreover, the difference between PS and MH is enlarged, and the value
of PS is even more emphasized when the number of destinations
increases. In other words, forcing the relay to decode all messages
becomes more costly when there are more destinations.
Fig. 5a shows the -outage rate region for R1 and R2 for a xed
minimum common outage probability of 0.01, Pavg = 1 dB, d = 0.3,
d1 = 0.2, d2 = 0.4 and = 4 for N = 2. It can be seen that MH and MHLC
achieve a much larger -outage rate region with respect to DT and
the -outage rate regions achievable with PS and PSLC are very close
to the upper bound. Therefore, allowing the relay to help each of
the destinations individually is almost optimal without the need of
link combination at the destinations, when the relay is close to the
source.
Allowing individual outage events brings in an extra degrees
of freedom into the system. Fig. 5b shows the amount of gains
obtained and compares -outage rate regions for DT and PSLC protocols under common and individual outage constraints. Under
individual outage setup, the -outage rate region is the set of all
target rates (R1 , . . ., RN ) such that a xed outage probability vector
(1 , . . ., N ) is satised. The gains are more emphasized for approximately equal target rates R1 and R2 , and diminish when either R1
or R2 is zero.
In Fig. 6a, we plot the expected values of the ratios


(PSLC)
P S1
(PSLC)
(PSLC)
P S1
+ P S2

(PSLC)
P R1
(PSLC)
(PSLC)
P R1
+ P R2

(35)

for Pavg = 0.5 dB, R1 = 1, R2 = 1, d1 = 0.5, and = 4 for N = 2. We observe

that, when D2 is colocated with D1 (d2 = 0.5), both the source
and the relay allot their power equally among the two users,
and E{} = E{} = 0.5. When 0.5 < d2 < 0.5, E{}, E{} > 0.5. This is
because the mean values for a2 and c2 are larger than the mean values for a1 and c1 . Therefore, both the source and the relay allocate
more power for D1 to decrease the common outage probability. The
opposite is true for d2 > 0.5. It can be seen that E{} and E{} reach
their maximum values for d2 = 0. In addition to these, we observe
that the relay location d has limited effect on E{} curve, while E{}
curve is highly dependent on the relay location.
When the system outage constraint is removed and individual
(i)
(i)
outage declarations are possible, a range of (Pout,1-min , Pout,2-min )
pairs become feasible for xed target rates R1 and R2 . Fig. 6b shows
(i)
(i)
the individual outage probability region Pout,2 vs Pout,1 for i = DT, PS,
PSLC and the lower bound for N = 2. It is seen that PSLC improves
upon the whole region PS attains. Moreover, PSLC is very close to
(i)
(i)
the lower bound for a range of (Pout,1 , Pout,2 ) around (0.01, 0.01).
5. Conclusion
In this paper, we study the broadcast relay channel in terms
of probability of outage under long-term power constraint and
delay-limited transmission. We propose ve different protocols,
direct transmission, multihop, multihop with link combination,
path selection and path selection with link combination. We analyze both system outage and individual outage events and nd
minimum common outage probability and the individual outage probability region. We also nd -outage rate regions for all
protocols. Finally, for comparison we write upper bounds on the
achievable rates in the broadcast relay channel and derive a lower

707

bound on outage probability and an upper bound on -outage rate

region. Our results indicate that enforcing the relay to help all destinations simultaneously is limiting the system performance. Partial
decoding brings in signicant gains in the BRC setting.
Acknowledgements
This material is based upon work supported by the Scientic and
Technological Research Council of Turkey, TUBITAK, under Grant
No. 112E059. The material in this paper was presented in part at the
7th International Wireless Communications and Mobile Computing
Conference, Istanbul, Turkey, July 2011.
References
[1] Kramer G, Gastpar M, Gupta P. Cooperative strategies and capacity theorems for relay networks. IEEE IT 2005;51(9):303763,
http://dx.doi.org/10.1109/TIT.2005.853304.
[2] Bhaskaran S. Gaussian degraded relay broadcast channel. IEEE Trans Inf Theory
2008;54(8):3699709, http://dx.doi.org/10.1109/TIT.2008.926306.
[3] Behboodi A, Piantanida P. Cooperative strategies for simultaneous and
broadcast relay channels. IEEE Trans Inf Theory 2013;59(3):141743,
http://dx.doi.org/10.1109/TIT.2012.2226699.
[4] Chen L. On rate region bounds of broadcast relay channels. In: Proceedings
of 44th conference on information sciences and systems. 2010. p. 16,
http://dx.doi.org/10.1109/CISS.2010.5464881.
[5] Liang Y, Veeravalli V. The impact of relaying on the capacity of broadcast channels. In: Proceedings of IEEE international symposium on information theory.
2004. p. 403, http://dx.doi.org/10.1109/ISIT.2004.1365440.
[6] Zhao L, Chung S-Y. Marton-Marton coding for a broadcast relay network. In:
Proceedings of IEEE international symposium on information theory. 2013. p.
12826, http://dx.doi.org/10.1109/ISIT.2013.6620433.
[7] Yildirim SA, Yuksel M. Multiple description coding based compressand-forward for the broadcast relay channel. In: Proceedings of IEEE
international symposium on information theory. 2012. p. 1948,
http://dx.doi.org/10.1109/ISIT.2012.6283639.
[8] Yildirim SA, Yuksel M. Cooperative broadcasting with successive renement
based compression. In: Proceedings of IEEE international symposium on information theory. 2013. p. 29804, http://dx.doi.org/10.1109/ISIT.2013.6620772.
[9] 3GPP further advancements for eutra: physical layer aspects. TR 36.814 v2.0.1
Technical Specication Group Radio Access Network Release 9; March 2010.
Available fromhttp://www.3gpp.org/DynaReport/36814.htm
[10] Yang Y, Hu H, Xu J, Mao G. Relay technologies for wimax and
lte-advanced mobile systems. Commun Mag IEEE 2009;47(10):1005,
http://dx.doi.org/10.1109/MCOM.2009.5273815.
[11] Tse D, Viswanath P. Fundamentals of wireless communication. Cambridge University Press; 2005.
[12] Caire G, Taricco G, Biglieri E. Optimum power control over fading channels. IEEE
Trans Inf Theory 1999;45(5):146889, http://dx.doi.org/10.1109/18.771147.
[13] Li L, Goldsmith A. Capacity and optimal resource allocation for fading broadcast
channels Part II: Outage capacity. IEEE Trans Inf Theory 2001;47(3):110327,
http://dx.doi.org/10.1109/18.915667.
[14] Hasna M, Alouini M-S. Optimal power allocation for relayed transmissions over
Rayleigh-fading channels. IEEE Trans Wirel Commun 2004;3(6):19992004,
http://dx.doi.org/10.1109/VETECS.2003.1208833.
[15] Yuksel M, Yuksekkaya B. Minimum outage probability for the relay channel
under individual power constraints. IEEE Commun Lett 2012;16(7):9825.,
http://dx.doi.org/10.1109/LCOMM.2012.050912.112192.
[16] Gunduz
D,
Erkip
E.
Opportunistic
cooperation
by
dynamic
resource allocation. IEEE Trans Wirel Commun 2007;6(4):144654,
http://dx.doi.org/10.1109/TWC.2007.348341.
[17] Reznik A, Kulkarni SR, Verd S. Broadcast-relay channel: capacity region
bounds. In: Proceedings of IEEE international symposium on information theory. 2005. p. 8204, http://dx.doi.org/10.1109/ISIT.2005.1523451.
[18] Liang Y, Veeravalli VV. Cooperative relay broadcast channels. IEEE Trans Inf
Theory 2007;53(3):90028, http://dx.doi.org/10.1109/WIRLES.2005.1549626.
[19] Isikman A, Yuksel M. Minimum common outage probability for the
broadcast relay channel. In: Proceedings of 7th international wireless communications and mobile computing conference. 2011. p. 8949,
http://dx.doi.org/10.1109/IWCMC.2011.5982665.
[20] Cover TM, Thomas JA. Elements of information theory (Wiley series in telecommunications and signal processing). Wiley-Interscience; 2006.
[21] Tse D. Optimal power allocation over parallel Gaussian broadcast channels. In:
Proceedings of IEEE international symposium on information theory. 1997. p.
27, http://dx.doi.org/10.1109/ISIT.1997.612942.
[22] Cover TM, Gamal AE. Capacity theorems for the relay channel. IEEE Trans Inf
Theory 1979;25(5):57284, http://dx.doi.org/10.1109/TIT.1979.1056084.